Device and method for image-based testing of a cable harness

Abstract

This document proposes a device and a method for inspecting an object. It also proposes a device and a method for teaching and / or training a machine learning algorithm. An embodiment of the device (100) for inspecting an object (O) comprises: at least one camera (10) arranged and configured to capture an object (O) located in a field of view (20) of the at least one camera (10) from a first perspective; and at least one reflective element (30) arranged and configured in the field of view (20) of the at least one camera (10) to provide the at least one camera (10) with a second perspective of the object (O). The at least one camera (10) is further configured to generate image data (12) of the field of view (20) and send it to an evaluation unit (40).The evaluation unit (40) is set up to receive the image data (12) and to determine the current state of the object (O) from the image data (12).

Description

[0001] This document proposes a device and a method for inspecting an object. It also proposes a device and a method for teaching and / or training a machine learning algorithm.

[0002] To manufacture cable harnesses, various cables are connected on a mounting board according to assembly instructions. The individual cable ends are connected with a plug or crimped onto a common contact. Due to the variability of different cable harness designs, the manufacturing process is largely manual. Manually performed processes are inherently prone to errors. Accordingly, manual processes require intensive testing of the result, in this case the manufactured cable harness, to prevent subsequent failures of a means of transport, such as a vehicle or aircraft, in which the manually manufactured cable harness is installed. Automated testing of the cable harness arranged on the mounting board cannot be performed from multiple, especially all, sides without installing additional testing equipment.

[0003] US patent 2013 100 277 A1 discloses a device and a method for inspecting a wiring harness. Specifically, it discloses a workbench for the manufacture or testing of electrical wiring harnesses, which provides the technician with the documentation for the manufacture or testing of electrical wiring harnesses in a digital format and in an interactive manner.

[0004] From DE 10 2015 112 722 B4 a device is known which uses a camera to detect cable harnesses.

[0005] A device for the visual inspection of a cable harness is known from KR 10 164 513 3 B1. The device for the visual inspection of a cable harness has a connector receptacle into which a connector of a cable harness is inserted. The connector receptacle is provided with a mirrored surface. A reflection of the cable harness in the connector receptacle is detected by a camera positioned above the receptacle.

[0006] A system for inspecting a wiring harness is known from JP 2019 060 685 A. The wiring harness is arranged on a mounting board. Mirrors are arranged on the mounting board. A movable camera travels along the mounting board and the wiring harness, capturing parts of the wiring harness via the mirrors or via the camera's own perspective.

[0007] A system for the visual inspection of a wiring harness, based on the use of artificial intelligence, is known from KR 10 2378186 B1. The system has a first camera positioned above an illuminated mounting board. The wiring harness is arranged on the mounting board. A second camera is positioned on one side of the mounting board and captures a connector of the wiring harness.

[0008] However, existing testing methods are inadequate. In particular, the information typically obtained is insufficient to guarantee a reliable test.

[0009] Therefore, there is a need for improved testing concepts. In particular, there is a need for testing concepts that ensure reliable testing with the simplest possible setup. Furthermore, there is a need for testing concepts that deliver continuously improved, and especially more accurate, results.

[0010] For this purpose, a device according to claim 1 and a method according to claim 5 are proposed.

[0011] According to a first aspect, a device for inspecting an object is proposed. The inspection can be non-contact and / or image-based. In this case, the device can also be referred to as a device for non-contact and / or image-based inspection of an object. The device comprises at least one camera. The at least one camera is arranged and configured to capture an object located within the field of view of the at least one camera from a first perspective. The device comprises at least one reflective element. The reflective element is arranged within the field of view. The reflective element is arranged and configured to provide the at least one camera with a second perspective of the object. The at least one camera is further configured to generate image data of the field of view. The device comprises an evaluation unit.The camera is set up to send image data to the evaluation unit. The evaluation unit is set up to receive the image data and determine the current state of the object from the image data.

[0012] The device offers the advantage of easily acquiring additional information about the object, for example, information about the object not only from a first perspective but also from a second perspective. This additional information enables improved inspection, particularly improved defect detection. Adding another camera in a different position would result in a more complex device. The effort and / or cost of additional cameras that could capture the object from another perspective can be saved. In other words, with a single camera, information about the object, such as surfaces, sides, or parts of the object, can be optically captured and provided that would normally be inaccessible to the camera, for example, from the camera's own perspective or facing away from its field of view.

[0013] The object's current state can contain information about whether the object is functioning correctly or not, for example, whether the current state corresponds to a planned or desired target state. The object's current state can also contain information about its orientation, position, and / or type. The object's perspective can be captured directly by the at least one camera. The object's second perspective can be captured indirectly by the at least one camera, for example, through reflection from the reflective element. The device can capture a multitude of images of the object from the first perspective using the at least one camera. The device can also obtain a multitude of images of the object from the second perspective using the at least one camera and the at least one reflective element.Taking numerous photos from the first perspective and / or the second perspective increases the accuracy of the examination.

[0014] For example, the current state can provide information about the orientation and / or type of object. This current state can then be compared to a desired target state. For instance, the comparison can determine whether the object is in a desired orientation, or, in other words, whether the determined current state represents a desired orientation of the object, or whether the determined current state contains a desired orientation of the object. Additionally or alternatively, the comparison can determine what type of object is located at the investigated site, or, in other words, whether the determined current state represents a desired type of object, or whether the determined current state contains a desired type of object.

[0015] The evaluation unit can be a computing unit, either as part of the device or arranged within the device. At least a part of the evaluation unit, for example, the entire evaluation unit, can be designed as a separate entity from the device, such as an entity separate from the other elements of the device or located in a different location. For example, the evaluation unit can include a server unit, such as a cloud server unit, or be designed as a server unit, such as a cloud server unit.

[0016] According to one embodiment, the evaluation unit can be configured to extract partial image data from the image data, particularly by means of a preprocessing stage. The partial image data can, for example, correspond to the object or contain information about the object. The partial image data can correspond to the object. Additionally or alternatively, the partial image data can correspond to the first perspective and / or the second perspective. The evaluation unit can be configured, particularly by means of a verification stage, to determine the current state of the object from the partial image data. The evaluation unit can be configured, particularly by means of the verification stage, to assign the current state to one of at least two predefined classes, for example, those that have been learned and / or trained.

[0017] In this way, the evaluation unit provides efficient, simple and / or accurate processing or evaluation of the recorded information.

[0018] The device can, for example, include at least one camera. In other words, the at least one camera can be designed as a single camera. The at least one camera has a field of view. The at least one camera can be a high-resolution camera or be designed as a high-resolution camera. If two or more cameras are provided in the device, the two or more cameras can be arranged in different positions to capture the object from different perspectives. This further increases the accuracy of the inspection.

[0019] The device can have at least one additional reflective element. This additional reflective element can be positioned within the field of view and provide a third perspective of the object. The camera can have a fixed relative position to the object.

[0020] The evaluation unit can have a preprocessing stage and a verification stage. The preprocessing stage can be configured to receive the image data, extract the partial image data, and send it to the verification stage. The verification stage can be configured to receive the partial image data, determine the current state of the object, and assign the current state to one of at least two predefined classes.

[0021] Determining the current state and / or assigning the current state to a class can be performed using a machine learning algorithm, in particular one that has been trained and / or learned. The test stage can be designed as a machine learning algorithm that needs to be trained and / or learned, especially during the first operating phase of the device. The test stage can also be designed as a machine learning algorithm that has been trained and / or learned, especially during a second operating phase of the device, or it can already include such a learning algorithm. In other words, a predefined machine learning algorithm, for example, one that has been trained and / or learned, can be implemented in the evaluation unit. The machine learning algorithm can, for example, be trained and / or learned using a device according to a second aspect, which will be described later.

[0022] The additional perspective of the object obtained through the reflective element, and the resulting additional information about the object, allows an AI model (AI: Artificial Intelligence) to be trained with more partial image data and / or partial image data with at least one additional perspective. This can improve inspection, for example, by making defect detection performed by the AI ​​model more accurate. For instance, a defect can be detected with a higher probability.

[0023] The evaluation unit can be configured to receive, in particular, synthesized training partial image data, especially via a first interface. The testing stage can be configured, in particular via a first interface, to receive, in particular, synthesized training partial image data.

[0024] In an initial operating phase of the device, the evaluation unit can be configured to monitor, train, and / or instruct the machine learning algorithm using, in particular, synthesized learning partial image data and / or learning objects. The evaluation unit can be trained to determine into which of at least two, for example, exactly two, classes the learning partial image data and / or the learning objects are to be categorized. In this way, the evaluation unit can define at least two, for example, exactly two, classes into which the learning partial image data and / or the learning objects are to be categorized.The evaluation unit can be set up in a second operating phase of the device to determine the current state of the object using the learned and / or trained machine learning algorithm and to assign it to one of at least two predefined classes.

[0025] The object is designed as a cable harness arranged on a mounting board. The device can therefore be used for image-based inspection of the cable harness arranged on a mounting board. The at least one camera is arranged and configured to capture a cable harness located within its field of view from a first perspective. The at least one camera is arranged and configured to capture the entire cable harness located within its field of view from a first perspective. The at least one reflective element is arranged and configured to provide the at least one camera with a second perspective of at least a part or section of the cable harness.The system may include at least one further reflective element which may be arranged in the field of view and may provide a third perspective of at least one part or section or a second perspective of another part or section of the wiring harness.

[0026] The evaluation unit is designed to extract partial image data from the received image data. This partial image data corresponds to at least one part or section of the wiring harness. The partial image data can correspond to one and / or each of the numerous parts or sections and the provided perspective(s).

[0027] The evaluation unit is configured to determine the current state of at least one part or section of the wiring harness, for example, the entire wiring harness, or a multitude of parts or sections, from the partial image data and to assign the respective current state to one of at least two predefined classes, such as those that have been learned and / or trained. The evaluation unit can include a preprocessing stage and a verification stage. The preprocessing stage can be configured to receive the image data, extract the partial image data, and send it to the verification stage. The verification stage can be configured to receive the partial image data. The preprocessing stage can automatically extract the partial image data using image processing algorithms and / or markings on the assembly board.The extraction can be performed via previously manually defined image areas that correspond to at least one part or section of the cable harness or the multitude of parts or sections of the cable harness.

[0028] Determining the current state and assigning the current state to a class can be carried out using the testing stage, in particular a trained and / or learned machine learning algorithm.

[0029] The evaluation unit can be configured to receive, in particular, synthesized learning partial image data. In an initial operational phase of the system, the evaluation unit can be configured to monitor, train, or instruct the machine learning algorithm using the, in particular, synthesized learning partial image data and / or learning wiring harnesses. In this way, the evaluation unit can learn or train into which of at least two classes the learning partial image data and / or the learning wiring harnesses should be classified.

[0030] In other words, during the initial operating phase of the device, the evaluation unit can be configured to monitor, train, or instruct the machine learning algorithm using, in particular, synthesized learning partial image data and / or learning cable harnesses. This allows the evaluation unit to define at least two classes into which the learning partial image data and / or the learning cable harnesses are to be categorized.

[0031] The evaluation unit can be set up in a second operating phase of the system, using, for example, a learned and / or trained machine learning algorithm to assign an actual state of the cable harness / part of the cable harness to one of at least two classes.

[0032] The at least one camera can have a fixed position relative to the wiring harness and / or the mounting board. The device can have a plurality of reflective elements. The reflective elements can be arranged within the field of view. At least one of the plurality of reflective elements can be located within the field of view of at least one of the at least one camera, for example, a plurality of cameras. Each reflective element can be associated with a part or section of a plurality of parts or sections of the wiring harness. Each of the reflective elements can be arranged and configured to provide the at least one camera with a second perspective of its associated part or section of the wiring harness.In addition to the multitude of reflective elements, the device may have at least one further reflective element which may be arranged in the field of view and may provide a third perspective from one of the multitude of parts or sections.

[0033] The at least one reflective element can be designed as a mirror, in particular as a magnifying mirror. The at least one reflective element can be arranged on the mounting board. In one embodiment, the at least one reflective element can, for example, be connected to the mounting board as a mirror. In this case, the mirror can be designed as an element separate from the mounting board. In a second embodiment, a surface of the mounting board can be mirrored, at least in part, and the mirrored section of the surface can form the at least one reflective element.

[0034] The at least one reflective element can have a plane of reflection, in particular a mirror plane, which is inclined relative to the mounting board, in particular by 16°, 18°, 20°, 22° or 24°. Inclinations between 10° and 30° are generally conceivable. The at least one camera can have a focus area, in particular a static / fixed one, which includes the at least one reflective element and the wiring harness. In other words, the at least one camera can have a static / fixed focus area that sharply images the wiring harness and the reflective element(s).

[0035] Using magnifying mirrors allows for a better view of details or subsections of parts or sections of the wiring harness. For example, using magnifying mirrors can make it easier to see the edges of the mounting board that are outside the focus range of at least one camera. In other words, the magnifying mirrors allow for a more detailed view of parts or sections of the wiring harness, such as edges that are outside the focus range of at least one camera.

[0036] The part or section of the wiring harness can include at least one connector and / or at least one mating connector and / or other elements, such as fasteners, etc. The entire mounting board can, for example, be positioned within the field of view of one or more cameras. Alternatively, only a section of the mounting board can be positioned within the field of view of one or more cameras.

[0037] For example, a section or part of the wiring harness may have a multitude of connectors and / or mating connectors. Accordingly, several, for example, all, of the multitude of connectors and / or mating connectors can be checked. In particular, the actual state of several, for example, all, of the multitude of connectors and / or mating connectors can be checked and assigned to a class, e.g., a status. For example, the determined state can be output for several, in particular, all, of the multitude of connectors. For example, an orientation and / or a position and / or a type of several, e.g., all, of the multitude of connectors and / or mating connectors can be determined and compared with a respective target state. For each of the several of the multitude of connectors and / or mating connectors, a result of the comparison can be determined and output.

[0038] According to a second aspect, a device for image-based instruction and / or training of a machine learning algorithm is proposed. The machine learning algorithm can be taught or trained using one or more of a multitude of learning objects. The device includes at least one camera. This camera is configured to capture, for example, one of a multitude of learning objects sequentially from a first perspective. The one learning object from the multitude of learning objects is positioned within the field of view. The device also includes at least one reflective element. This reflective element is positioned within the field of view and is configured to provide the camera with a second perspective of the one learning object from the multitude.

[0039] The at least one camera is positioned and configured to generate image data of the field of view. The device includes an evaluation unit. The at least one camera is configured to send the image data to the evaluation unit. The evaluation unit is configured to receive the image data and feed it to the machine learning algorithm for training purposes.

[0040] According to one embodiment, the evaluation unit can be configured to receive image data and extract sub-image data from it. This sub-image data can correspond to one of a multitude of learning objects. The evaluation unit can then be configured to feed this sub-image data to the machine learning algorithm for training purposes.

[0041] The device may include at least one additional reflective element, which may be positioned within the field of view. This additional reflective element may be arranged and configured to provide a third perspective of a learning object.

[0042] In general, the device according to the second aspect can be constructed at least almost identically to the device according to the first aspect. The device according to the first aspect and the device according to the second aspect may differ, for example, in the implementation of the evaluation unit, but otherwise be identical. The device according to the first aspect and the device according to the second aspect can be combined into a single device whose structure may, for example, correspond exactly to that of the device according to the first aspect. With such a combined device, the evaluation unit can, for example, first be trained or educated, as described in relation to the device according to the first aspect and the device according to the second aspect.The trained evaluation unit can then be used, as described in relation to the device according to the first aspect, to test an object, for example a cable harness.

[0043] At least one camera can have a fixed position relative to the learning object. Each camera can have a single field of view. The learning object can be a cable harness mounted on a board, which in this context can be referred to as a learning board or learning cable harness.

[0044] The device according to the second aspect can capture at least one further learning object from the multitude of learning objects and evaluate the captured information. For example, the device according to the second aspect can perform capture and evaluation as many times as necessary until all of the multitude of learning objects have been captured and evaluated one or more times. In this way, the machine learning algorithm of the evaluation unit can be further trained, and any subsequent testing using the evaluation unit can be improved.

[0045] The device can capture information about the orientation, position, and / or type of the learning object and use it to train the machine learning algorithm. For example, the learning object can be captured in different states, and the information about these states can be fed to the machine learning algorithm. To perform improved training, multiple states of the learning object can be detected and used for training. In this way, the model being trained can learn and subsequently recognize the classes or states.

[0046] Referring to the exemplary design of the learning object as a learning cable harness, the numerous connectors and / or mating connectors of the cable harness can be recorded and fed to the machine learning algorithm. For example, the connectors and / or mating connectors can be recorded in different states and fed to the machine learning algorithm, and / or different types of connectors and / or mating connectors can be recorded and fed to the machine learning algorithm.

[0047] According to a third aspect, a method for examining an object is proposed. The method comprises providing at least one camera. The method comprises capturing, using the at least one camera, an object arranged in the field of view of the at least one camera from a first perspective. The method comprises providing at least one reflective element, which is or will be arranged in the field of view. The method comprises providing, using the at least one reflective element, a second perspective of the object. The method comprises generating, using the at least one camera, image data of the field of view. The method comprises transmitting, using the at least one camera, the image data to an evaluation unit. The method comprises receiving the image data using the evaluation unit.The procedure involves determining, using the evaluation unit, the current state of the object from the image data.

[0048] The determination step involves extracting, using the evaluation unit, sub-image data from the image data that corresponds to the object. The current state determination step involves determining, using the evaluation unit, the object's current state from the sub-image data. The current state determination step involves assigning, using the evaluation unit, the current state to one of at least two predefined classes, for example, learned and / or trained classes.

[0049] The object is designed as a cable harness. Accordingly, the acquisition step comprises capturing, using at least one camera, a cable harness positioned within the field of view from a first perspective. The extraction step can include extracting, using the evaluation unit, from the image data, partial image data that corresponds to at least one part or section of the cable harness and / or the first perspective of the object and / or the second perspective of the object. The determination of the current state comprises determining, using the evaluation unit, the current state of at least one part or section of the cable harness from the partial image data. The assignment step comprises assigning, using the evaluation unit, the current state to one of at least two predefined classes.

[0050] The details described in relation to the device according to the first aspect can be implemented accordingly in the method according to the third aspect, and vice versa.

[0051] According to a fourth aspect, a method for teaching and / or training a machine learning algorithm is proposed. The method can be performed on one or more of a multitude of learning objects. The method includes providing at least one camera. The method includes capturing, using the at least one camera, a learning object from the multitude of learning objects located within the camera's field of view from a first perspective. The method includes providing at least one reflective element, which is or will be located within the field of view. The method includes providing, using the reflective element, a second perspective on one of the multitude of learning objects. The method includes generating image data of the field of view using the at least one camera. The method includes transmitting the image data to an evaluation unit using the at least one camera.The process involves receiving image data via the evaluation unit. The process also includes feeding the image data to the machine learning algorithm in order to train the machine learning algorithm.

[0052] The process can include training the machine learning algorithm. The process can include extracting, using the evaluation unit, sub-image data from the image data that corresponds to one of the many learning objects. The process can include feeding the sub-image data to the machine learning algorithm. The process can include training the machine learning algorithm.

[0053] The procedure may involve repeating one or more of the previous steps for each additional learning object from the multitude of learning objects.

[0054] The details described in relation to the device according to the second aspect can be implemented accordingly in the method according to the fourth aspect and vice versa.

[0055] A fifth aspect relates to a computer program with program code means which, when loaded into or running on a computer or processor (e.g., a microprocessor, microcontroller, or digital signal processor (DSP)), causes the computer or processor (e.g., microprocessor, microcontroller, or DSP) to execute one or more steps, or all steps, of the process steps previously described with respect to the evaluation unit of the device according to the first and / or third aspect, and / or with respect to the process steps described according to the second and / or fourth aspect. Furthermore, a program storage medium or computer program product containing the aforementioned computer program is provided.

[0056] Furthermore, for example, the computer program according to the fifth aspect can be stored in the evaluation unit of the device according to the first aspect and / or according to the third aspect and cause the evaluation unit to execute one or more or all of the aspects and / or steps of the procedure described above with respect to the evaluation unit according to the second and / or the fourth aspect. Furthermore, for example, the computer program according to the fifth aspect can be stored in the evaluation unit of the device according to the first aspect and / or according to the third aspect and cause the evaluation unit to execute one or more or all of the aspects and / or features described above with respect to the evaluation unit.

[0057] Further features, properties, advantages and possible variations will become clear to a specialist from the descriptions below, which refer to the attached drawings. Fig. Figure 1 shows a schematic representation of a device for image-based inspection of an object; Fig. Figure 2 shows a schematic representation of an evaluation unit and a camera of the device. Fig. 1 for image-based inspection of an object; Fig. Figure 3 shows a schematic representation of a device for testing a cable harness arranged on a mounting board; Fig. Figure 4 schematically shows an evaluation unit and a camera of the device. Fig. 3 for testing a cable harness arranged on a mounting board, Fig. Figure 5 shows a version of the design from a user's perspective; and Fig. Figure 6 shows a variant design from the perspective of a camera used to inspect a cable harness arranged on a mounting board.

[0058] Specific details are set forth below, without limitation, to provide a complete understanding of the present disclosure. However, it is clear to a person skilled in the art that the present disclosure can be used in other embodiments that may differ from the details set forth below. For example, specific configurations and embodiments are described below, which are not to be considered limiting.

[0059] It is clear to those skilled in the art that the explanations set forth below may be implemented using hardware circuits, software means, or a combination thereof. The software means may be related to programmed microprocessors, artificial intelligence or a general-purpose computer, an ASIC (Application-Specific Integrated Circuit), and / or DSPs (Digital Signal Processors). It is also clear that even if the following details are described in relation to a method, these details may also be implemented in a suitable device unit, a computer processor, or memory connected to a processor, the memory containing one or more programs that carry out the method when executed by the processor.

[0060] The Fig. Figure 1 shows a schematic representation of a device 100 for image-based inspection of an object O. The device has a camera 10 in whose field of view 20 an object O is positioned. The object O, here exemplified as a cuboid, has several sides, of which the front face OF and the top face OT are marked with the corresponding reference symbols. To distinguish the two sides from each other, an oval is drawn on the front face OF and a triangle on the top face OT. Due to the position of the camera 10, the camera 10 can only capture the front face OF of the object O as a first perspective. The other sides are not captured by the camera 10. Accordingly, the camera 10 perceives the object O from a first perspective 21.

[0061] A reflective element 30 is arranged in the field of view 20 of the camera 10. In the exemplary representation of the device 100, this element is designed as a mirror 30. The mirror 30 has a reflective surface that provides the camera 10 with a second perspective 22 of the top side OT of the object O. The mirror 30 thus allows a view of sides of the object O that are inaccessible to the camera 10 due to its relative position to the object O. The camera 10 generates image data of the field of view 20, in which the object O and, for example, the mirror 30 are arranged. The image data therefore contains the first perspective 21, or self-perspective, of the camera 10 on the object O and the second perspective 22, which is provided by the mirror 30.

[0062] The device further comprises an evaluation unit 40, which is connected to the camera. The image data 12 are transmitted from the camera 10 to the evaluation unit 40.

[0063] The Fig. Figure 2 shows a schematic representation of the evaluation unit 40 and the camera 10 of the device 100 for image-based inspection of an object. The evaluation unit 40 has a preprocessing stage 42 and an inspection stage 44. The Fig. The image data 12 described in section 1, which is sent to the evaluation unit 40, is received by the evaluation unit 40, in particular by the preprocessing stage 42. In the left part of the Fig. An arrow begins at point 2 between camera 10 and evaluation unit 40 and points to a drawing that illustrates the image data 12. The image data 12 corresponds to the field of view 20 of camera 10 and the object O and mirror 30 located within it. More precisely, camera 10 can capture the front OF of object O from the first perspective 21 and the top OT from the second perspective 22, using mirror 30.

[0064] The image data 12 are sent to the evaluation unit 40 and to a preprocessing stage 42 located therein. The preprocessing stage 42 extracts from the image data 12 the parts 120, 122 of the image data that correspond to object O and, from both perspectives 21, 22. This is in the lower right part of the Fig. Figure 2 shows the image to which the arrow points, starting between preprocessing stage 42 and inspection stage 44. Both partial image data sets 120 and 122 show the relevant image data of object O, specifically the top side OT and the front side OF. In the example shown, both partial image data sets 120 and 122 are transmitted to or sent to inspection stage 44, which previously has, for example, a user-defined target state of object O for comparison. For instance, a target state could be that the same features are printed on the front and top sides OF and OT of object O. In this example, it is assumed that the target state should represent an oval on the front and top sides OF and OT of object O. The inspection algorithm in inspection stage 44 determines the actual state of the object from the received partial image data sets 120 and 122 and compares this actual state with the target state.Furthermore, test stage 44 is designed to indicate whether the target state O1 is present or not O2. In the present case, the actual state differs from the target state, so that the evaluation unit 40, in particular test stage 44, indicates via O2 that the actual state does not correspond to the target state.

[0065] The S1 interface allows target state partial image data to be sent to the evaluation unit 40, in particular to the test stage 44, from which a target state is determined and compared with an actual state to be determined from the partial image data 120, 122.

[0066] The Fig. Figure 3 shows a device 200 for testing a cable harness KB arranged on a mounting board B. The mounting board B can also be referred to herein as the cable harness board or simply as the cable board or board. The device 200 can be attached to the device 100 from Fig. 1 correspond, so that the in relation to Fig. 1 and Fig. The details described in section 2 can be implemented accordingly in the device 200.

[0067] System 200 includes a camera 210, within whose field of view 220 a cable harness KB with first and second connector ends KB1, KB2 is arranged. Due to the position of camera 210, it can only capture a top view of the cable harness KB and the connector ends KB1, KB2 from a first perspective. One or more side and / or bottom views, or other parts or sections of the cable harness KB, are not captured by camera 210. Accordingly, camera 210 perceives the cable harness KB from a first perspective.

[0068] In the field of view 220 of the camera 210, two reflective elements 230, 232 are arranged as examples, which are in the Fig. 3. Examples are provided as two mirrors 230, 232. Alternatively, only one mirror can be used, as in relation to Fig. 1 described or more than two mirrors may be provided. The mirrors 230, 232 each have a reflective surface that provides / enables the camera a second perspective 222a, 222b of the two connector ends KB1, KB2. The mirrors 230, 232 thus each enable a second perspective 222a, 222b on sides (from the viewpoint of the camera 210) of the wiring harness KB that are inaccessible to the camera 210 due to its relative position to the wiring harness KB. The camera 210 generates image data 212 of the field of view 220 in which the wiring harness KB and the mirrors 230, 232 are arranged. The image data 212 thus contains the first perspective, or self-perspective, of the camera 210 on the wiring harness KB and the respective second perspective 222a, 222b on the connector ends KB1, KB2.

[0069] The device 200 further comprises an evaluation unit 240, which is connected to the camera 210. The image data 212 are transmitted from the camera 210 to the evaluation unit 240.

[0070] The evaluation unit 240 can be used by the evaluation unit 40. Fig. 2 correspond, so that the in relation to Fig. 1 and Fig. The details described in section 2 can be implemented accordingly in evaluation unit 240.

[0071] The in relation to the Fig. The image data 212 described in section 3, which are sent to the evaluation unit 240, are processed in the Fig. 4 received from the evaluation unit 240, in particular from the preprocessing stage 242. In the left part of the Fig. 4 begins an arrow between camera 210 and evaluation unit 240 and points to a part of the Fig. Figure 4 illustrates the image data 212. The image data 212 corresponds to the field of view 220 of the camera 210 and the cable harness KB and the mirrors 230, 232 arranged therein. More precisely, the camera 210 can capture the top view from the first perspective and the respective side view (from the camera's point of view) of the connector ends KB1, KB2 through the mirrors 230, 232 from the second perspective 222a, 22c. The image data 212 is sent to the evaluation unit 240 and to a preprocessing stage 242 arranged therein.

[0072] The preprocessing stage 242 extracts from the image data 212 the parts 212a-c or sections of the image data that correspond to the two connector ends KB1, KB2 and the perspectives. This is in the lower right part of the Fig. Figure 4 shows the point indicated by the arrow that starts between preprocessing stage 242 and inspection stage 244. All partial image data 212a-c show the relevant parts of the image data of the cable harness KB, more precisely the top view (from the perspective of camera 210) and the side view (from the perspective of camera 210) of the connector ends KB1 and KB2.

[0073] In the Fig. In the four exemplary examples shown, all partial image data 212a-c for each of the connector ends KB1 and KB2 are sent to a test stage 244, which, for example, has a previously trained machine learning algorithm. The two connector ends KB1 and KB2 can belong to the same or different connector types. The trained machine learning algorithm serves as the basis for checking the partial image data 212a-c transmitted for each connector end. For example, a target state could be that the connector ends KB1, KB2 are correctly assembled. In other words, a connector end can have a connector holder / mother connector and a connector, and the target state describes the correct positioning / placement of the connector in the corresponding connector holder / mother connector. Additionally or alternatively, a target state could be that the correct connector ends KB1, KB2, e.g.,The connector ends of a correct connector type, or of several connector types, are mounted. Additionally or alternatively, a desired state may be that the connector ends KB1 and KB2 are mounted in the correct location / position. Additionally or alternatively, a desired state may be that the connector ends KB1 and KB2 are mounted in the correct orientation.

[0074] The trained machine learning algorithm in test stage 244 determines the actual state of each connector end KB1 from the received partial image data 212a-c and compares it to the corresponding target state. Test stage 244 is configured to indicate whether the target state is OK (=OK) or not OK (=NOK). In this example, for connector end KB2, the connector was not correctly positioned in the connector holder / mother connector. The machine learning algorithm registers, for example, the number of cables to connector end KB2 (four, for example) and the missing cable end in the side view (from the camera's perspective) from mirror 232. From this, the machine learning algorithm determines the actual state of connector end KB2 and detects a deviation from the target state.

[0075] Evaluation unit 240, specifically test stage 244, displays NOK (=not OK) as the evaluation result for connector end KB2, meaning that connector KB2 is not correctly positioned, for example. Furthermore, evaluation unit 240, specifically test stage 244, displays OK (=OK) as the evaluation result for connector end KB1, meaning that connector KB1 is correctly positioned, for example. Additionally or alternatively, evaluation unit 240 can display NOK as the evaluation result for connector end KB1, meaning that the wrong connector type was used, for example. Additionally or alternatively, evaluation unit 240 can display OK as the evaluation result for connector end KB2, meaning that the correct connector type was used, for example.Additionally or alternatively, evaluation unit 240 can display "NOK" as the evaluation result for connector end KB1, meaning, for example, that connector end KB1 was not installed in the correct orientation. Additionally or alternatively, evaluation unit 240 can display "OK" as the evaluation result for connector end KB2, meaning, for example, that connector end KB2 was installed in the correct orientation. All these possibilities are purely illustrative and can be combined as desired to obtain an evaluation result of "OK" or "NOK". For example, two or more conditions, such as all conditions, can be met to obtain an evaluation result of "OK". This means, for example, that connector end KB1 of the correct connector type must be mounted in the correct orientation and in the correct location to receive an evaluation result of "OK". If a condition is not met, "NOK" can be displayed.

[0076] During an initial operating phase of the device 200, training partial image data can be sent via interface S21 to the evaluation unit 240, particularly to test stage 244. This data is used to train the machine learning algorithm. The training partial image data can consist of connector-type-dependent partial image data from various perspectives of the respective connector type. Alternatively or additionally, during an initial operating phase, training cable harnesses corresponding to a target state can be sequentially arranged in the device 200, and extracted partial image data from these harnesses can be used to train the machine learning algorithm.

[0077] Fig. Figure 5 shows an exemplary view of a mounting board B from a user's perspective. The cable harness KB is arranged on the mounting board B and spaced apart from it by cable harness holders KBK. Each of the six ends of the cable harness KB is fixed in a cable harness holder KBK. Relative to the ends and a cable section, reflective elements, in this case mirrors 231, 233, 235, 237, 238, and 239, are arranged. From a user's perspective, mirrors 231, 233, 235, 237, and 239 allow a side view of the cable harness ends. From a user's perspective, mirror 238 allows a rear view of a cable section. Alternatively, instead of the mirrors shown in the illustration, a surface of the mounting board B could also be mirrored to provide the corresponding perspectives.

[0078] Fig.Figure 6 shows an exemplary view from the perspective of a camera used to inspect a cable harness mounted on a mounting board. A cable harness end with a connector KB1 is visible. The cable harness end is secured by a cable harness holder KBK. Therefore, the camera has a top view of the cable harness end and connector KB1 from a first perspective. A mirror 236 positioned next to the cable harness end KB1 provides the camera with a second perspective of the cable harness end and connector KB1. In this case, from the camera's perspective, this second perspective is a side view of connector KB1.

[0079] The described configurations provide an improved, in particular more accurate and adaptive, testing concept.

Claims

[1] Device (100; 200) for testing a cable harness (KB) arranged on a mounting board (B), comprising: - at least one camera (210) which is in a fixed relative position to the mounting board, arranged and configured to completely capture a cable harness (KB) arranged in a field of view (220) of the at least one camera (210) from a first perspective, - at least one reflective element (230, 232) arranged in the field of view (220) of the at least one camera (210), which is arranged and configured to provide the at least one camera (210) with a second perspective of at least part of the cable harness (KB), wherein the at least one camera (210) is configured to generate image data (212) of the field of view (220) and to send it to an evaluation unit (240), wherein the evaluation unit (240) is configured: - to receive the image data (212), - extracting from the image data (212) partial image data (212a-c) which correspond to at least one part of the cable harness (KB) of the first perspective and the second perspective, and - to determine the actual state of at least one part of the wiring harness (KB) from the partial image data (212a-c) and to assign the actual state to one of at least two predefined classes (OK, NOK). [2] Device (200) according to claim 1, wherein the evaluation unit (240) is configured, in the first operating phase of the device (200), to learn / train a machine learning algorithm (244) with, in particular synthesized, learning partial image data and / or learning cable harnesses in order to learn and / or train the machine learning algorithm (244) into which of at least two classes (OK, NOK) the learning partial image data and / or the learning cable harnesses are to be divided, and in a second operating phase of the device (200) to assign the actual state of at least one part of the cable harness (KB) to one of the at least two classes (OK, NOK) by means of the learned and / or trained machine learning algorithm (244). [3] Device (200) according to one of claims 1 to 2, wherein: - the at least one reflective element (230, 240) is designed as a mirror, in particular as a magnifying mirror, and is arranged on the mounting board (B), in particular inclined by 16°, 18°, 20°, 22° or 24° relative to the mounting board (B), and / or - the at least one camera (210) has a, in particular static or fixed, focus area which includes the at least one reflective element (230, 232) and the cable harness (KB). [4] Device (200) according to one of claims 1 to 3, wherein the at least one part of the wiring harness (KB) comprises at least one connector (KB1, KB2) and / or at least one wiring harness section and / or at least one section of the mounting board (B) within the field of view (220) of the at least one camera (210). [5] A method for testing a wiring harness (KB) arranged on a mounting board (B), the method comprising: - Providing at least one camera (210) arranged in a fixed relative position to the mounting board; - completely capturing, by means of the at least one camera (210), a cable harness (KB) arranged in a field of view (220) of the at least one camera (210) from a first perspective; - providing at least one reflective element (230, 232) which is or will be arranged in the field of view (220) of the camera (210); - providing, by means of the at least one reflective element (230, 232), a second perspective of at least a part of the cable harness (KB); - generating, by means of the at least one camera (210), image data (212) of the field of view (220); - sending, by means of the at least one camera (210), the image data (212) to an evaluation unit (240); - receiving the image data (212) by means of the evaluation unit (240); and - extracting, by means of the evaluation unit (240), from the image data (212), partial image data (212a-212c) which correspond to the at least part of the cable harness of the first and second perspectives; - Determining, by means of the evaluation unit (240), the actual state of at least one part of the cable harness (KB) from the partial image data (212a-212c); and - Assigning, by means of the evaluation unit (240), the actual state of at least one part of the wiring harness to one of at least two predefined, in particular learned and / or trained, classes (OK, NOK).

Images